Abstract:The rotational speed of a rotating machine in industrial applications often varies over time. Examples include the motor of an elevator, gearbox of a wind turbine and motor of a robot arm. Speed variation induces additional amplitude modulation (AM) effects to vibration signals, and such AM effects overlap with those induced by faults. It is necessary to normalize vibration signals, i.e., accurately estimate the speed-variation-induced AM effects and sufficiently remove them from vibration signals, otherwise would result in missing alarms and/or false alerts in fault detection. Existing methods for estimating the speed-variation-induced AM effects are either not able to grasp all the effects such as resonance property or have the fault-induced AM effects included. To address this problem, this paper proposes a new method for estimating the speedvariation-induced AM effects through a piecewise power fitting (PPF) manner. The proposed method can well capture the resonance property without being influenced by the fault signatures. The estimated speed-variation-induced AM effects are then removed from vibration signals through a normalization method. The normalized vibration signals are employed by a deep autoencoder (AE) for intelligent fault detection. The effectiveness of the proposed PPF method and its contribution to intelligent fault detection is validated with two experimental datasets and compared with several state-of-the-art methods.

Normalizing Vibration Signals with a Novel Piecewise Power Fitting Method for Intelligent Fault Detection of Rotating Machinery